# LlamaIndex LlamaIndex is the data framework for LLM applications — born (as "GPT Index") from the question that defined [[Retrieval-Augmented Generation (RAG)]]: how do you get *your* data into a model's context, usefully? It provides the pipeline pieces: ingestion connectors for every data source, indexing structures, retrieval strategies, and query engines, in Python and TypeScript. Open source at the core. Where LangChain generalized into "framework for everything LLM," LlamaIndex kept its center of gravity on **data-to-LLM**: parsing, chunking, indexing, retrieving. Its agent and workflow abstractions exist, but retrieval quality remains the reason to pick it. ## The product stack | Layer | What | Cost | |---|---|---| | LlamaIndex (OSS) | The framework itself | Free | | [[liteparse]] | Local document parser — spatial extraction, complexity routing | Free (Apache 2.0) | | LlamaParse | Cloud parser for hard documents (dense tables, charts, scans) | Paid | | LlamaCloud | Managed ingestion/retrieval pipeline | Paid | The liteparse/LlamaParse split is the honest version of open-core: the local tool tells you outright when a document is complex enough to justify the paid parser — and its complexity-detection step is exactly the routing signal. ## References - [LlamaIndex Website](https://www.llamaindex.ai) - [LlamaIndex on GitHub](https://github.com/run-llama/llama_index) - [Documentation](https://docs.llamaindex.ai) ## Related - [[Retrieval-Augmented Generation (RAG)]] — the pattern it industrialized - [[liteparse]] — its local parsing layer - [[Vector Store]] · [[Embeddings]] — the machinery it orchestrates